Meas. Sci. Technol. 11 (2000) 887–901. Printed in the UK PII: S0957-0233(00)08350-8

Applications of planar laser inducedfluorescence in turbulent reactingflows

A Cessou†, U Meier‡ and D Stepowski†§

† Universite de Rouen, UMR-CNRS 6614/CORIA, Place Emile Blondel,76821 Mont Saint Aignan, France‡ DLR, Institut fur Verbrennungstechnik, Pfaffenwald Ring 38-40, 70569 Stuttgart, Germany

E-mail: denis.stepowski@coria.fr

Received 4 October 1999, accepted for publication 21 January 2000

Abstract. The basic principles of laser induced fluorescence are described with progressivepresentation of simple and multi-level interaction schemas providing the relationship betweenthe fluorescence signal and the local concentration of the investigated species in the laserfield. Essential differences between laser induced fluorescence and flame chemiluminescenceemissions are given. Techniques for imaging in the planar LIF configuration and proceduresfor calibration of the fluorescence intensity in absolute concentration values are described.Then, examples of imaging applications are given for instantaneous mapping of OH, CH, O2

and temperature fields in turbulent reacting flows. In spite of uncertainties mainly due tocollisional quenching effects, these imaging experiments are useful to interpret the interactionprocesses involved in turbulent combustion especially when several quantities can besimultaneously or conditionally registered.

Keywords: laser diagnostic, turbulent combustion, laser induced fluorescence, planarimaging, laser spectoscopy, laser thermometry

(Some figures in this article are in colour only in the electronic version; see www.iop.org)

1. Introduction

Laser induced fluorescence (LIF) is the spontaneous isotropicemission of light by which the molecules of a given speciesthat have been selectively driven onto an excited electronicstate by tuned laser excitation (optical pumping) relax totheir ground state. The fluorescence power is then directlyproportional to the excited state population through theEinstein probability coefficient A for spontaneous emission.In hot reacting media collisions and chemical reactions canalso populate excited states, but the excited populations andthe subsequent emissions induced by these processes aremuch lower than those induced by pulsed optical pumping.

Absorption of the laser photons by molecules is directlyresponsible for the population of the excited state in thelaser field which is a thin sheet of light in the planarlaser induced fluorescence (PLIF) technique. Besidesrelaxation by spontaneous emission of fluorescence at rateA, other de-excitation processes such as stimulated emission,collisional quenching, energy transfers or predissociations(at global rate Q in a simplified two-level schema) arecompetitively involved in the interaction. The dynamicsof the population transfer must be carefully examined toobtain the concentration of the investigated species in the

§ Corresponding author.

excited state as a function of its global concentration. Thenit is made possible to derive that global concentration fromthe measured intensity of the laser induced fluorescenceemission.

Emphasis will be put on the simple regimes for whichthe fluorescence emission is locally proportional to the laserirradiance and to the molecular population in its lowerstate. Different calibration procedures may be used todetermine the proportionality coefficient in order to deriveabsolute concentration data from measured fluorescenceintensity. In many cases a simple reference sample ofthe investigated molecular species is not currently available(radical species. . .) and the calibration must be performedin a particular zone of a reacting flow where the absoluteconcentration of the molecule has been calculated or canbe measured by another technique such as absorptionspectroscopy.

Complementary to fluorescence emission, absorption ofthe laser excitation light is involved in the interaction process.From an energy balance, the fluorescence emitted over aspatial domain D of the laser field can be considered as apartial recovery of the small laser energy that is absorbedby the molecules over that domain D and over the spectralwidth of the absorption line. This energy recovery is partialwith the Stern–Volmer factor, A/(A + Q), that accounts forthe part A which is radiatively relaxed, relative to the total

0957-0233/00/070887+15$30.00 © 2000 IOP Publishing Ltd 887

A Cessou et al

Figure 1. Two-level schema.

de-excitation rate A + Q. Thus in situ calibration of theglobal efficiency of a fluorescence system (operating in thelinear regime) can be made with the line of sight absorptionmethod by measuring the attenuation of the same laser overthe absorption linewidth and over an enlarged optical pathwhere the investigated concentration is homogeneous or hasa known relative profile. This complementary point of viewshows up that the laser irradiance may suffer significantattenuation when progressing in the absorbing medium whichin turn would make the local level of fluorescence emissionbe lower than expected. Great care must be taken to minimizesuch a bias with a proper pumping configuration. Thesame kind of bias may occur if the emitted fluorescencelight is significantly trapped by self-absorption before beingdetected; as well as for laser attenuation, we will describeconfigurations that can minimize fluorescence trapping.

When the three main problems (i.e. attenuation of thelaser irradiance, determination of the local fluorescenceefficiency, trapping of the emitted fluorescence) are more orless solved, one must examine how the specific populationprobed by the laser in a particular energy level of theground state can be related to the global population of theinvestigated molecule. In the ground state, the populationsare distibuted over the rotational sublevels according toBoltzmann statistics—which is not true in the upperstate where the distribution over the sublevels is stronglyperturbed by the laser pumping. Accordingly, rotationaltemperature can be derived from successive (with short delay)fluorescence measurements at different laser wavelengthsprobing the same molecule on different rotational sublevels ofits ground state. The sensitivity of this two-line thermometryis better when the probed sublevels are set wider apart oneach side of the maximum of the Boltzmann distribution.On the other hand, when performing a simple concentrationmeasurement that should be independent of temperature it isrecommended to probe the population of a sublevel close tothe maximum of the distribution.

This article will present examples of imagingapplications in turbulent non-premixed reacting flows byplanar laser induced fluorescence of OH, CH, toluene andO2. In many cases, maps of a single or simple scalarare not sufficient to provide a relevant interpretation of anexperiment. Therefore the described applications are oftenperformed with simultaneous imaging of two quantities, OHand CH, OH and velocity, fuel/air ratio, which is fruitful

to investigate the interacting processes involved in turbulentcombustion.

2. Principles of LIF interaction

This section presents the bases of laser induced fluorescence[1] in order to put in evidence its potential and its limitationsas a diagnostic method. The medium is schematicallydescribed by energy levels over which the population of theinvestigated species is distributed, and the action of the tunedlaser pulse is described as a perturbation of the distributionusing detailed balance for the population transfers.

2.1. Two-level model

In this very simple model the medium is described by twoelectronic energy levels E1 and E2 over which the molecularconcentration N is distributed with concentrations N1 and N2

respectively (N1 + N2 = N ) as shown in figure 1. The laserradiation tuned on the transition frequency ν = (E2 −E1)/h

is characterized by its spectral energy density Uν whichquantifies the number density of photons with the suitablefrequency. Uν = I/cδν where I is the local irradiance ofthe laser in W cm−2, and δν is the spectral width of thelaser radiation as long as it is larger than that of the coveredabsorption line.

Several elementary processes can induce transitionsbetween the energy levels: collisions at rate N1Q12 populatethe upper level which is competitively depopulated bycollisional relaxation, called quenching, at rate N2Q21, andby spontaneous emission of isotropic light with frequency ν

at rate N2A. A is a spectroscopic constant of the investigatedmolecule whereas Q terms due to collisions with othermolecular species are given by Q = ∑

i σi〈Vi〉Ni . σi is thespecific cross section of collision partner i, Vi is the relativevelocity that scales as

√T and Ni is the concentration of

partner i. Therefore Q terms are linearly increasing withpressure, and at atmospheric pressure Q21 is much higherthan A.

Absorption at rate N1B12Uν populates the upper levelwhich is simultaneously depopulated by stimulated emissionof coherent light in the laser direction at rate N2B21Uν . In asimple model with two single states B12 = B21 = B.

888

Applications of PLIF

At the initial condition where Uν = 0, the steady statebalance equation is

dN2

dt= N1Q12 − N2(Q21 + A) = 0 (1)

which yields

N2/N1 = Q12/(Q21 + A) ≈ Q12/Q21. (2)

The corresponding Boltzmann equilibrium statistics gives

N2/N1 = exp[−(E2 − E1)/kT ]. (3)

For a typical electronic transition in the UV around 300 nm,and at a typical flame temperature around 1700 K, one obtainsQ12/Q21 ≈ 10−3, and N2 is negligible. Thus for the initialcondition one can set N2 = 0, and N1 = N .

When the medium is submitted to tuned laser irradiance,the detailed balance equation is

dN2

dt= N1(BUν + Q12) − N2(BUν + A + Q21). (4)

As BUν � Q12, and Q21 � A, the excited concentrationincreases according to

N2(t) = NBUν

2BUν + Q[1 − exp −(2BUν + Q)t] (5)

where Q is a shortcut expression for Q21. The first factorgives the excited concentration at steady state which isreached after a short delay called the pumping time, τ =1/(2BUν +Q). At atmospheric and higher pressure τ is muchshorter than the duration of the laser pulse (�t � 10 ns), andN2(t) follows a quasi-steady state evolution as a function ofUν(t) during the laser excitation.

At any instant the rate of spontaneous emission offluorescence photons per unit volume is given by dn/dt =N2A, such that the power of the quasi-steady statefluorescence emission is

P(t) = NBUν(t)

2BUν(t) + QAV hν

�

4π(6)

where V is the interaction volume, and � is the solid anglein which the fluorescence is emitted. Thus the fluorescenceemission is proportional to the number of molecule NV inthe probed volume V which is determinated by the imagedsection of the laser field, but it may still depend on theconcentration of other species involved in the collisionalquenching Q.

To overcome this quenching dependence it has beenproposed [2] to saturate the transition with a very stronglaser pulse (BUν � Q) such that the strong fluorescencepower would become P = NAV hν�/8π independent ofquenching and laser power. Unfortunately this ideal fullsaturated regime is not actually achievable [3] (especiallyover a laser sheet as in planar LIF), and partial saturation (i.e.non-linearity with respect to Uν) has more drawbacks thanadvantages.

In the regime of linear laser excitation where Q12 �BUν � Q, the power of the fluorescence emission reducesto

P(t) = NBP0(t)A

Q�L

�

4π

hν

cδν(7)

where �L is the length of the section of the laser field fromwhich the fluorescence is emitted, and P0(t) is the powerof the laser over that section. The linearity with the laserpower is an essential and useful feature of this regime [4]because the fluorescence emission is then independent of thefocusing degree of the laser (the transverse thickness of thelaser is difficult to measure and to keep constant). It allowsfor simple calibration and referencing procedures as will bedescribed in the next section. However one must insure thatthe local laser power P0 has not been significantly attenuatedby absorption during its propagation up to the probed zone.

In low pressure flames where the pumping time can belonger than the duration of the laser pulse (1/Q > �t),the excited concentration increases linearly with time duringthe laser excitation up to a peak level which is independentof collisional quenching, N2(�t) = NBUν�t . This peakis followed by an exponential decay of N2(t), and of thesubsequent fluorescence, at rate Q [5]. This particular timeresolved regime of LIF has been useful to obtain quenchingdata and to perform parametric studies of its global behaviourin low pressure premixed flames as a function of air/fuel ratioand reaction progress [6].

2.2. Multi-level models

Actually each electronic state is composed of severalvibrational levels which are themselves composed ofrotational sublevels. These sublevels are coupled bycollisional energy transfer processes, and the gound statepopulation is distributed over the sublevels according to theequilibrium Boltzmann statistics. A hundred collisions arerequired to reach equilibrium over the vibrational structurewhile few collisions insure equilibrium among the rotationalsublevels. In a flame, around 1700 K, Boltzmann statisticsindicates that the equilibrium population of the lowervibrational level v = 0 is about 95% of the total. Thus, in afirst step, we will only consider the fundamental vibrationalstates as illustrated in figure 2.

The laser is tuned on a particular line which couplesa rotational level J ′′ of the ground state with a rotationallevel J ′ of the excited electronic state. Within the groundstate the relative population of the J ′′ absorbing level isstill determined by the Boltzmann equilibrium distributionduring the optical pumping in linear regime as insured bythe strong collisional coupling (rotational energy transfer)between the rotational sublevels. Within the exited state theexcess population brought by the laser onto the particularsublevel J ′ is progressively redistributed by rotational energytransfer over the neighbour sublevels, but competitively allthe upper population is quenched toward the ground state.Therefore the relative population among the upper sublevelscannot reach equilibrium Boltzmann distribution during thelaser excitation since the rate of rotational energy transfer isonly a few times higher than the rate of electronic quenching(and both are linearly increasing with pressure in the sameway).

In the regime of linear laser excitation where BUν � Q,the evolution of the global excited concentration follows

d∑

NJ ′

dt= NJ ′′BUν −

∑NJ ′QJ ′ . (8)

889

A Cessou et al

Figure 2. Multi-level schema.

As long as the relaxation rates AJ ′ and QJ ′ are not toodependent on J ′ over the range covered by rotational energytransfer, one can define and use mean relaxation rates A andQ over this range. Then, provided that the fluorescenceis spectrally collected over a large band issuing from allthe excited sublevels [7], the global fluorescence intensityis the same as for a two-level model except that it is nowproportional to the concentrationNJ ′′ on the particular groundstate sublevel probed by the laser.

The relative population of that particular sublevel isgiven by the Boltzmann equilibrium distribution that prevailson the ground state

NJ ′′

N≡ 2J ′′ + 1

kTexp

(−EJ ′′

kT

). (9)

To have a fluorescence intensity proportional to the totalconcentration and independent of temperature it is thennecessary to pump a J ′′ level for which the relative populationis nearly constant over the expected temperature range. Thisis obtained when EJ ′′ ≈ kT . For example when measuringOH with a laser tuned on the Q18 line, the ratio N8/N variesby less than 5% over the temperature range 1300–2000 K [8].On the other hand the relative population of that rotationallevel J ′′ = 8, is close to the maximum of the distributionand one must take care that absorption does not significantlyattenuate the laser power.

In an opposite strategy, the Boltzmann distibutionover the ground state can be used to determine the localtemperature of the medium. LIF of a given species isthen performed at different wavelengths to probe its relativepopulation on different rotational sublevels of the groundstate. An example of this local thermometric technique withtwo lines of OH will be given in the next.

2.3. Chemiluminescence

In non-sooting flame natural UV–visible emission, calledchemiluminescence, is the spontaneous emission fromradical species (OH, CH, C2, . . .) which have been partiallycreated by chemical reaction in an excited electronic state

(in weak proportions and for complex reasons). If weconsider such an elementary reaction (beyond many others),as A + B → C, there is a generally low probability γ to haveA + B → C∗. C∗ is created on a particular sublevel of itsexcited state and, as previously explained, has no time to beequitably redistributed over the neighbour sublevels becauseof the fast electronic quenching toward the ground state. Abalance shows (as was previously made with optical pumpinginstead of chemical reaction) that the steady state value of thechemiluminescence intensity scales as C∗ = γ AB/Q. Thisintensity is then a qualitative measure of the conjunctionof the parent species A and B, rather than an estimateof concentration C. When radical species C is chemicallycreated it can diffuse or be convected downstream beforerecombining, whereas its excited fraction C∗ is quenched tothe ground state before it can be significantly transported.Therefore the chemiluminescence is constricted at the birthplace of C—the reaction zone. Instead LIF can probetransported C at any location in the laser field. This essentialdifference between the two processes is illustrated in figure 3.

3. Measurement techniques by planar LIF

3.1. Examples of suitable interaction schemas

Here we present spectral arrangements for laser excitationand subsequent fluorescence detection that have proven to beefficient to minimize some of the difficulties met in the laserinduced fluorescence method. The OH radical will be takenas typical example of a molecular species to be probed andimaged by PLIF in turbulent reacting flow.

A first technique, schematized in figure 4, is to exciteOH through the weak v′′ = 0, v′ = 1 band (λ = 285 nm).The laser attenuation is then negligible while the probedpopulation of the v′′ = 0 level is still a very high andsignificant fraction of the total population. The fluorescenceis then detected around 313 nm through the strong v′ =1, v′′ = 1 band of emission whose trapping is weak asthe absorbing population on v′′ = 1 is very low [9–11].This spectral configuration is useful and very efficient formapping experiments because elastic scattering noise at the

890

Applications of PLIF

Figure 3. Comparative schemas for chemiluminescence (top) and laser induced fluorescence (bottom).

Figure 4. Interaction schema (0 → 1), (1 → 0) for OH LIF.

laser wavelength can be easily rejected by placing a properbandwidth or high pass filter before the camera. Howeversome fluorescence emitted through the P branch of the (0,0)band via vibrational energy transfer may be detected withinthe bandwidth of the filter [12].

Another technique schematized in figure 5 is to pumpOH through the weak v′′ = 0, v′ = 3 band (λ ≈ 248 nm)into the v′ = 3 predissociative level of the excited electronicstate whose life time is ten times shorter than the collisionalquenching time at atmospheric pressure. The collisionfree fluorescence emitted through the (3,2) band is detectedaround 297 nm. This configuration has many advantages: thefluorescence is linear and basically insensitive to collisionssince the collisional quenching rate Q is dominated by theconstant rate of predissociation P which is ten times higher;on the other hand the fluorescence efficiency is then tentimes lower, and this must be compensated by using a stronglaser irradiance [13]. As in the previous (0,1)–(1,1) schema,laser attenuation by absorption and fluorescence trappingare negligible for the same reasons. However the requiredstrong UV laser induces significant oxygen fluorescence inthe lean high temperature zones of the reacting flow [14]. Theresulting interferences around 297 nm must be subtractedby monitoring the O2 fluorescence at another wavelength(337 nm). As will be seen later, this predissociativefluorescence schema applies also for oxygen when excitedin the UV around 193 nm.

The same kind of strong non-collisional transfer in theexcited state also occurrs for acetone. It can be used withsimilar benefit to excite the blue fluorescence of gaseousacetone (as a tracer) with a UV laser around 275 nm [15].

3.2. Planar imaging systems

Pulsed laser sources are directly available with excimer lasersthat provide suitable pulsed UV radiations (�t ≈ 25 ns) forexcitation of predissociated OH with KrF, or predissociatedO2 with ArF. The second harmonic of a pulsed YAG laser at532 nm is used to excite a tunable dye laser in the visible.This dye laser beam is then frequency doubled to generatethe tunable pulsed radiation (�t ≈ 8 ns) for UV laserexcitation of OH fluorescence. To excite the fluorescenceof CH which has very low absorption in the blue around430 nm, a tunable laser with high pulsed energy is required.In order to avoid optical breakdown induced by too highirradiance, use is made of a long duration (�t ≈ 1.5 µs)flash-lamp-pumped dye laser (Diana system of the SandiaLaboratory). In any case the spectral width of the tunablelaser is made significantly larger than that of the absorptionline to minimize effects of spectral drift of the dye laser fromshot to shot.

A typical arrangement for planar LIF is schematizedin figure 6: a system of cylindrical and spherical lenses isused to focus the laser beam onto a sheet of light that istypically 0.3 mm thick by 4 cm high. The fluorescenceemission is collected at 90◦ of the laser sheet, and imagedon a gated intensified ccd camera with typically a 100 mm–f/4 objective lens. A spectral pass-band filter is interposedin front of the camera to reject elastic scattering at the laserwavelength and to select the suitable fluorescence. In the UV,butylacetate liquid filter may be used as a high pass to rejectthe laser radiation with very high efficiency [13]. A typicalcamera format is a 500 × 500 array, 25 µ × 25 µ pixels.

A diode array may be used to measure the distributionof laser irradiance over the height of the light sheet inorder to correct for non-uniform laser excitation (flat fieldingprocedure) as well as to account for drift and fluctuation ofthe laser power. A trigger system is required to drive thepulsed laser and the synchronous amplification of the ccdover a suitable gate duration. This temporal discriminationimproves the rejection of flame luminosity that was partlymade by spectral filtering.

891

A Cessou et al

Figure 5. Interaction schema (0 → 3), (3 → 2) for predissociated OH LIF.

Figure 6. Typical arrangement for planar LIF.

3.3. Calibrations

In the regime of linear laser excitation, the quasi-steady statefluorescence signal is proportional to the laser power and tothe local population of the investigated molecule in the lasersheet. But it is also inversely proportional to the local rate ofelectronic quenching which has to be known as a function ofthe different collision partners. Thus calibrations are requiredto investigate the local influence of this quenching on theconcentration measurement.

The quenching rate can be calculated in various flamesusing available data for the specific quenching cross sectionsand weighting by the local mole fractions of these differentcollision partners [16, 17]. Another approach is to performdirect determination of the local effective quenching rate inlow pressure premixed flames by measuring the decay rateof the time resolved fluorescence signals. Both experimentsand calculations in laminar premixed flames show that theeffective quenching rate of OH and CH is nearly constantover the flames [6, 18], which indicates that quenching ispractically insensitive to the degree of reaction progressfor a given mixture. In addition equilibrium calculationsof OH and CH quenching in non-premixed flame showlow variations over the range of mixture fraction wherea significant concentration of OH or CH is predicted[19, 20]. However OH is often found in superequilibriumconcentration and over a broader range of mixture fraction

(especially on the fuel rich side) than calculated with theequilibrium assumption.

In any case a global calibration of the imaging system(including laser spectral energy, effective probed volume,solid angle of fluorescence detection, sensitivity of thedetection system,. . .) must be achieved to derive absoluteconcentration data from the local level of fluorescence signal.This is usually performed with the same imaging system ina particular zone of a simple reacting flow (often in the burntgases of a premixed flame) where the absolute concentrationof the investigated molecule can be calculated [20] or canbe measured by another technique such as line of sightabsorption spectroscopy [21].

Calibration by absorption spectroscopy can be per-formed in situ with the LIF imaging system by using laserinduced fluorescence as a local non-intrusive spectrometerwhich probes the laser irradiance over the absorption lineprovided that the absorbing concentration is constant. Sucha technique has been used in a laminar diffusion flame wherethe radial profiles of imaged OH fluorescence show two peakscorresponding to the two intersections of the conical flameby the laser. As shown in figure 7 the first peak (on the laserside) was always slighly higher than the second one. The rel-ative difference between the two peak areas is then a directmeasure of the optical depth due to OH absorption over thethickness of the first flame front. This technique provides theabsolute OH concentration integrated over a peak, which can

892

Applications of PLIF

Figure 7. Asymmetry of OH fluorescence peaks in a diffusionflame.

be used to calibrate the radial profiles of fluorescence signalin profiles of absolute OH concentration.

4. Examples of applications

4.1. OH and CH maps in jet flame

Simultaneous mapping of OH and CH fluorescence inturbulent jet flames of methane has been performed at SandiaLaboratories by Schefer et al [11, 22]. Two pulsed laserbeams (λ1 = 431.5 nm, �t1 = 1.8 µs; λ2 = 285.4 nm,�t2 = 8 ns) were combined into a common optical axis by adichroic mirror and formed into an overlapping sheet (0.3 mmthick) by a multipass cell. The laser sheet at λ1 excites the(v′′ = 0, v′ = 0) band of the A 2�–X 2% system of CH. Thefluorescence emitted at 90◦ to the laser sheet via the (v′ = 0,v′′ = 1) transition is selected by a 10 nm bandwidth filtercentred at 489 nm and focused onto a gated intensified ccdcamera. The CH4 Raman scattering that is simultaneouslyinduced near 489 nm by excitation at λ1 is also detected bythis camera. This Raman signal could be rejected using apolarizer but, as its contribution to the total signal is lessthan 10% in the reacting zone, it can be simultaneously usedwithout ambiguity to quantify the concentration of CH4 inthe fuel-rich region.

The laser sheet at λ2 excites the Q18 line of the (v′′ = 0,v′ = 1) band of the A 2&, X 2% system of OH. The (1,1) bandfluorescence emitted at 90◦ to the laser sheet in the oppositedirection is selected by a 10 nm bandwidth filter centred at312 nm and focused onto a second camera.

This arrangement provided a spatial resolution of0.4 mm × 0.4 mm over a field of 47 mm long × 15 mmhigh. The two opposite imaging systems were aligned usingan object frame in the test section. Both laser attenuationand fluorescence trapping were less than 5%. The CH4

Raman data were normalized by the number density of puremethane at room temperature. In this study the CH–CH4

system was sometimes operated without a filter to observethe strong CH resonance fluorescence signal along withthe Rayleigh scattering signal. This last strategy providedqualitative information on density variation in the flame. TheOH fluorescence system was calibrated using a flat flameburner, but no attempt was made to derive absolute CHconcentrations. The simultaneous images as illustrated infigure 8 show that, while the fuel consumption zone markedby CH is thin in all the flames investigated, the zones of highOH concentration that mark the region where H2 and COburnout occurs with radical recombination, is often broadenough to have internal turbulence structure, which cannotbe accounted for by the strained laminar flamelet approach.

4.2. OH and velocity maps in lifted flame

Simultaneous mapping of OH fluorescence and velocity fieldhas been performed in the stabilization region of a turbulentlifted jet flame by Cessou et al [23]. The objective ofthe experiment was to investigate the flow dynamics wherethe reaction takes place, and to determine the particularlocation where the instantaneous velocity must be measuredfor relevant comparison with the different models proposedto explain the stabilization process of the lifted jet flame[24–26].

Planar LIF of OH was excited over the light sheet(120 mm high × 0.4 mm thick) of a pulsed dye laser (�t =10 ns, F = 10 Hz, E = 20 mJ/pulse) tuned on the Q16line of the (0,1) vibrational band of the (X 2%–A 2&) systemat 283.9 nm. The fluorescence was imaged onto a gatedintensified ccd camera (576×384×14 bit, 50 ns gate) with a105 mm, f/4.5 objective lens preceded by a pass-band filterselecting the fluorescence from the (1,1) band around 315 nm.

The jet fluid and a slow air coflow were seeded with1 µ ZrO2 particles that are expected to mark the flowvelocity. Particle image velocity (PIV) was performed bycross correlation of two successive pictures obtained byimaging elastic (Mie) scattering of a double-pulse laser sheet(3 cm high × 0.3 mm thick, λ = 532 nm, 10 µs delay). Thescattered light was collected at 90◦ to the sheet in the oppositedirection to the fluorescence detector, and focused on a ccdarray (736 × 576 pixels) through an interferencial filter. Thetwo successive images were respectively stored on the oddand even frames of the ccd. Each instantaneous velocityvector represents the ensemble average over a 2 mm×2 mm×0.3 mm cell. The UV laser sheet and the green laser sheetswere superposed on the same plane, and a timing monitortrigged the UV laser pulse between the green pulses.

Figure 9 shows the superposition of an instantaneousvelocity field and the simultaneous OH fluorescence image.In the stabilization region the structure of the OH field is quitecomplex with a very sharp edge in the inner part of its base.Experimental results of Schefer et al [27] show that a liftedflame can return to upstream location when it has been carrieddownstream by vortical structure; thus a premixed reactionstructure able to propagate must exist (as responsible for theflame stabilization) in the lower part of the reactive zone. Inaddition, the flame stabilized in a layer where the mixture is

893

A Cessou et al

Figure 8. Simultaneous pictures of CH, CH4, OH turbulent jet flame (from [11]).

Figure 9. Simultaneous pictures of OH and velocity fields in thestabilization region of a lifted jet flame [23].

stoichiometric [28]; therefore one must identify the locationof a stoichiometric premixed combustion structure on the OHfluorescence image to investigate the flow dynamics wherethe flame stabilizes.

In a premixed propagating flame the OH profile ischaracterized by a sharp increase on the fresh gas sidefollowed by a long trail in the burnt gas side due to slow

three-body recombination. In a diffusion flame the OHprofile is smoother and much more symmetric. Thereforeon the instantaneous pictures of OH the precise location G offlame stabilization has been determined where the gradientof OH is a maximum whereas in previous studies [29, 30] itwas merely determined at the bottom M of the flame contour(which contour was approximated by the limit of the Miescattering field). Accordingly the relevant flow velocity istaken as the projection of the velocity vector at G normal tothe envelope, whereas in previous studies it was merely takenas the axial velocity at the bottom M.

With this refined conditional measurement at Gthe statistical distribution of velocity spreads quitesymmetrically, with a most probable value close tothe laminar deflagration velocity (SL = 0.43 m s−1)of a premixed stoichiometric methane–air flame, whichcorresponds to a sustained flame. The occurrences ofnegative velocities corresponds to events for which thecombustible mixture is locally entrained into the jet by avortical structure.

With simple measurement of the axial velocitycomponent at the bottom of the OH contour, the velocitydistribution is asymmetric with a most probable value higherthan SL, and no occurrence of a negative value.

This example of PLIF application illustrates the greatbenefit that can be obtained by performing simultaneousplanar measurement of two quantities, allowing forinvestigation of conditional data, together with instantaneousgradients.

4.3. PLIF of fuel/air ratio in engine

In some particular cases the local composition dependence ofcollisional quenching involved in LIF can be a useful featureto map the relative composition field of several species in aturbulent mixing flow. LIF of toluene is an example of sucha capability.

894

Applications of PLIF

Figure 10. Maps of fuel/air ratio and velocity field for different crank angles in the compression stroke of an engine.

Toluene presents a large and homogeneous absorption

band in the UV which can be excited in linear mode by

a broad-band pulsed KrF excimer laser (λ ≈ 248 nm).

Fluorescence is then induced over a red-shifted emission band

(265–340 nm). The intensity of the steady state fluorescence

emission is basically proportional to the concentration of

probed toluene fuel [F], to the spectral laser irradiance Iν , and

to the Stern–Volmer efficiency factor A/(A′ + Q), where A′

is a spectroscopic constant including all the rates of intrinsic

relaxation of the excited fluorescent state, while Q is the

rate of collisional quenching. Experiments performed in a

cell [31] filled with various collisions partners (O2, H2O,

CO2, N2, . . .) at different partial pressures show that oxygen

is by far the most efficient quencher of laser excited toluene.

The intensity of the fluorescence emission can be expressed as

SF ≡ ABIν[F]

A′ + Q≡ ABIν[F]

A′ + α[O2]. (10)

Under high pressure conditions where A′ � Q, as reachedin the engine above 3 bar, the fluorescence signal is thenbasically a linear function of the local fuel/air ratio.

SF ≡ ABIν

[F]

[O2]. (11)

However this relationship is slightly affected by temperaturevariation since oxygen quenching efficiency α and tolueneabsorption are functions of the local temperature. Testsperformed in the compression stroke of an engine between300◦ crank angle (CA) and 350◦ CA showed that the relativeeffects of mean temperature increase with crank angle on

895

A Cessou et al

the mean fluorescence intensity resulted in a linear decay ofabout 1% per degree CA, allowing for mean correction ofthis temperature bias. Figure 10 shows examples of maps offuel/air ratios conditioned on crank angle in the combustionchamber of a spark-ignition engine during the compressionstroke [32]. The fuel was natural gas doped with toluenevapour. In addition the mean velocity field obtained by PIVconditioned on the same crank angles has been superimposedon each image.

Such conditional diagnostics may be helpful toinvestigate the effects of inhomogeneities of fuel/air rationear the spark plug on the cycle-to-cycle fuctuation of thecombustion especially when operating under a fuel-leanregime as required for low pollution.

4.4. Two-line OH temperature field

The fluorescence of a given species can be successivelyexcited by two laser pulses (with a short delay over which themedium stays frozen) of irradiances I1 and I2 tuned on tworotational lines of a given absorption band. The intensity ratioof the fluorescence signals is then related to the rotationaltemperature T according to the Boltzmann distribution thatprevails over the ground state:

SF1

SF2= B1iI1g1(2J1 + 1)

B2j I2g2(2J2 + 1)exp

(−E1 − E2

kT

). (12)

Here, J1 and J2 are the rotational quantum numbers ofsublevels 1 and 2 with energies E1 and E2 in the groundstate. B1i and B2j are the Einstein absorption coefficients for(1 → i) and (2 → j ) electronic transitions, g1 and g2 arethe spectral overlap integrals between absorption and laserprofiles, respectively.

This expression implies several assumptions: thetransmission characteristics of the two detection systems haveto be identical, the laser excitations are not saturated and thefluorescence quantum yields (in other words, the collisionalrates for electronic quenching) are identical for both excitedstates.

The expression above holds as well in a referencemedium with known temperature TR . The temperature cantherefore be obtained from a pair of two-line fluorescencemeasurements of the same species in the investigated mediumand in a reference flame with known temperature TR:

T =[

1

TR

− k ln R

E1 − E2

]−1

with R = SF1

SF2

SR2

SR1. (13)

Here, SF1, SF2, SR1, SR2 are the intensities of the normalizedfluorescence signals induced by lasers 1, and 2 in theinvestigated medium (subscript F) and reference flame(subscript R), respectively. Normalization is performed withrespect to laser intensity in each measurement. The use ofa reference flame reduces the systematic error that results ifthe fluorescence quantum yields for the two excitation linesare not identical as assumed in equation (12).

The use of OH radical as the probed species for planartwo-line LIF imaging of rotational temperature in enginehas been performed by Meier [33, 34]. OH presents onedrawback but has important advantages for LIF thermometry

Figure 11. Engine with optical access.

in high pressure reacting flows: temperature is accessibleonly above approximately 1500 K at high pressure in theflame front and post-flame where significant levels of OHfluorescence are detected. This lower limit results fromthe rapid decrease of OH concentration with temperature.The corresponding large variations of OH LIF signal withtemperature are higher than current detector technologycan handle. Preliminary experiments in iso-octane flamesshowed that when exciting OH fluorescence by differentrotational lines of the (0 → 1) absorption band around283 nm, the signal to noise ratio (estimated by laser detuning)was excellent showing no interference signal from otherspecies. Tests performed in a high pressure burner up to60 bar showed that the excitation spectra are still able toresolve the rotational structure of OH. This advantage isdue to the large line spacing of the lines and to their lowpressure broadening coefficient of around 0.08 cm−1 bar−1,depending on temperature and gas composition. Additionaltests performed in engine operating conditions showed thatattenuation of the laser irradiance due to OH absorption wasabout 6% when tuned on the Q18 line of the (0 → 1) bandwhich has a higher absorption coefficient than those of thelines used for thermometry.

To perform two-line PLIF in the engine fuelled with iso-octane, superposition of two laser beams with wavelengthsclose together onto a common axis was achieved by apolarizing beam splitter. The combined beam was formedinto a sheet of about 0.3 mm thickness which traversed thecylinder in a plane parallel to the piston surface 5 mm belowthe pentroof of the combustion chamber (see figure 11). Onthe detection side, the fluorescence was collected through aquartz window in the piston, passed over an inclined (45◦)mirror, colimated by a lens and then separated by a 50%beam splitter. Each of the split beams was then focused ona ccd camera with an achromatic UV lens (105 mm, f/2).

896

Applications of PLIF

Figure 12. Experimental arrangement for two-line PLIF.

Figure 13. Examples of OH temperature fields for different crank angles.

Interference filters centred at 312 nm were used to rejectscattered laser light and flame luminosity. The cameras werealigned carefully for identical field of view. Any residualmismatch due to misalignment or camera optics imperfectionwas removed by software superposition of the two images.The laser pulses were temporally separated by 500 ns, whichis much shorter than the time scale for the fastest transportprocesses in the flame and insures adequate time resolutionof the two fluorescence signals. The temporal gates of theintensifiers of the two cameras were adjusted such that eachcamera recorded only the fluorescence induced by one of thetwo pulsed lasers, respectively. Figure 12 shows a schematicoverview of the experimental arrangement.

Since models predicted peak temperatures above 2500 Kin the engine, a pair of lines with large energy spacing inthe ground state had to be selected. The pair P11/R114of the (0 → 1) band of the A–X system of OH with�E = 3800 cm−1 has been chosen. Both lines are spectrallyisolated even at pressure well above the engine peak pressure,which is an essential advantage.

Non-uniformity of laser irradiance over the light sheetwas measured and then corrected for on average basis: anaverage irradiance distribution was recorded for each lasersheet before and after a set of experiments by filling thecylinder homogeneously with acetone vapour, and recordingthe pictures of acetone fluorescence (above 400 nm) inducedby the same laser sheets as used for OH excitations. Thetemperature was determined from the intensity ratio of thefluorescences induced by the two lasers relative to the valueof this ratio in a reference flame. A premixed methane/airflame burning on a simple stainless steel tube (4 mm diameter)was placed inside the cylinder of the engine to keep allexperimental parameters constant. The temperature of thisreference flame was independently measured by CARSspectroscopy to be 2130 ± 50 K.

A series of 40 pairs of fluorescence images were recordedin the engine for various operating conditions. Figure 13shows typical instantaneous images of temperature field fordifferent crank angles. The engine speed was 1200 rpmwith ignition at 300 ◦CA. Later than 370 ◦CA the snr became

897

A Cessou et al

Figure 14. Interaction schema for broad-band LIF of O2.

too poor for reliable temperature measurement due to strongquenching with rising pressure. On the other hand, unburntgas regions where no OH is detected appear black on theseimages. It can be seen how the flame propagates from a regionnear the spark plug through the chamber with increasingcrank angle. For each image, a space average temperaturewas calculated over a region behind each flame front wherethe temperature is relatively homogeneous. The ensembleaverage of this temperaure increases significantly with crankangle from 2330 K at 342 ◦CA to 3015 K at 366 ◦CA, withcycle-to-cycle rms fluctuation about 200 K. However theaccuracy for the absolute temperature measurement is about130 K at 2200 K, and about 215 K at 2800 K. Such dataare useful in as much as the average flame speed can besimultaneously obtained as a function of the crank angleto investigate correlations between temperature, deflagrationvelocity and pressure in the engine.

4.5. PLIF of oxygen in turbulent O2/H2 jet flame

Planar LIF of oxygen has been performed by Stepowskiet al in a turbulent non-premixed O2/H2 flame using abroadband ArF excimer laser [35]. For a given oxygenconcentration laser absorption and subsequent fluorescenceemission are strongly increasing with the temperature. Thereverse configuration of the investigated flame with axialinjection of oxygen in a fast coflow of hydrogen is particularlywell suited to investigation by O2 LIF: the laser irradiancehas not been attenuated by cumulative absorption whenit reaches the reaction zone where the hot and weakamounts of O2 are detected with a maximum sensitivity.Assuming a fast H2–O2 reaction, the conditional rate of scalardissipation (then, the rate of oxygen consumption) can bederived from measurement of the fluorescence gradient atthe stoichiometric surface.

The rates of reaction in turbulent non-premixed flamesare strongly dependent on the mixing dynamics so long as thechemical kinetics are fast relative to the mixing. When themixing is dominated by turbulence all species and enthalpyare transported at the same rate, and all of the concentrationsand temperature can be related to a single conserved scalar,called the mixture fraction Z, wich is the local mass fractionof fluid originated in the fuel feed. This can be introduced

[36] in the species conservation equations to express thereaction rate in a simple form

ωi = −ρD|∇Z|2 ∂2Yi

∂Z2(14)

where ρ is the density, Yi is the mass fraction of species i

and D is the molecular diffusivity. This simple form of ωi

shows a mixing contribution through the scalar dissipationχ = 2D|∇Z|2, and a chemical contribution as ∂2Yi/∂Z2.If the chemistry can be represented by a fast single reactionthe reactants cannot coexist and ∂2Yi/∂Z2 is constricted inan infinitely thin reaction sheet, the stoichiometric surfaceZ = ZS . The average consumption rate of oxygen is thengiven by

ωO2 = ρS

2ZS

∫χP (χ, ZS) dχ. (15)

Thus even with a simplified schema of turbulent combustionknowledge of mean and fluctuation fields is not sufficientas the average rate of reaction depends on the conditioned(by Z = ZS) average dissipation noted in short cutχS . Use of coupling functions between species massfractions, temperature and mixture fraction in the flame withassumptions of high Damkohler number and unit Lewisnumber for a fast H2–O2 reaction constricted at Z = ZS =1/9 provides the relation between ∇Z and ∇|O2| aboutstoichiometric. Then measurement of the gradient of thefluorescence signal SIN at its inflection in the instantaneousthin reaction zone provides the conditioned dissipation rate

χS = 2DS

25βS

|∇SIN |2 (16)

where β is the global coefficient relating the fluorescencesignal to the local mole fraction of oxygen. β dependson experimental parameters (laser irradiance, detectionefficiency,. . .) and on the cross section σ(T ) for the O2–laser interaction. βS is the value of β at the stoichiometrictemperature (TS ≈ 2900 K) where the conditionalmeasurement of χS is made on each fluorescence picture.The next paragraph is devoted to the study of σ(T ) whenoxygen fluorescence is excited by a broad-band ArF laser[37].

As schematized in figure 14, many lines with absorptioncoefficients αij are reached by a broad-band ArF laser

898

Applications of PLIF

Figure 15. Absorption cross section of a broad-band ArF laserversus oxygen temperature.

(193.0 nm < λ < 193.8 nm) in the X 3&g → B 3&u

electronic transition system of oxygen. The excited statepopulation can be transferred to a predissociated state at rateP (which is an intrinsic constant of oxygen) or collisionallyquenched at rate Q, or radiatively relaxed by spontaneousemission of fluorescence (200–280 nm) at rate A. Theglobal fluorescence quantum yield A/(P + Q + A) is thenpractically constant (≈A/P ) as predissociation dominates allother deexcitation processes (P ≈ 1011 s−1) at atmosphericpressure. The fluorescence signal is given by

S ≡ IA

P[n0α04 + n1α15 + n2α26 + · · ·]N (17)

Figure 16. Example of single shot image of O2 PLIF in the turbulent jet flame.

where ni = Ni/N is the relative population of ground statesublevel i. Owing to the relative position of potential energycurves in X 3& and B 3& states and to the Franck–Condonprinciple for transition probability, the absorption coefficientsαij are strongly increasing with vibrational number i. Asthe relative populations ni of the sublevels with high i arealso increasing with temperature according to the Boltzmanndistribution law, the global absorption over all the lines in thelaser profile (as well of the subsequent fluorescence intensity)is an increasing function of the temperature. Figure 15 showsthe calculated evolution of the effective absorption crosssection σ(T ) for a broad-band ArF laser as a function ofthe oxygen temperature.

The spectrum of the emitted fluorescence contains anumber of lines (over a 80 nm range on the red sideof the laser wavelength) increasing with the temperature.But as the temperature increases the red spectral shift ofthe fluorescence decreases and the fluorescence spectrumhappens to overlap the laser frequency. The globaltransmission of the fluorescence, as it is selected by the high-pass filter rejecting the laser radiation, must be calculated asa function of the temperature, using a library of fluorescencespectra, to be finally accounted in the global sensitivity β(T )

of the LIF system.The experiment has been performed at ONERA

on the MASCOTTE test bench [38]. The coaxialinjector consisted of a central tube of 5 mm fed with

899

A Cessou et al

Figure 17. Maps of average reaction rate in the turbulent flame (over 500 shots).

liquid oxygen at 50 g l−1 surrounded by a parallel annularduct of 16 mm fed with hydrogen at 300 K and 15 g l−1

flow rate under atmospheric pressure. This injector waslocated at the base of a horizontal open chamber (55 mm ×55 mm×156 mm) equipped with four quartz windows cooledby helium flow film. The duration of each run, about oneminute, allowed registration of 500 single shot fluorescencepictures.

The pulsed radiation (λ = 193 nm, �λ = 0.8 nm,W = 170 mJ, �t = 15 ns, f = 10 Hz) of an ArF laser(Lambda Physik LPX100) was focused into a vertical lightsheet (100 mm high × 800 mm thick) passing through thejet axis. The focalization mounting was placed in a tube fedwith nitrogen to avoid laser absorption. The fluorescence wasimaged at 90◦ to the laser sheet onto a gated intensified ccdcamera (�t gate = 50 ns) with a 105 mm f/4.5 objectivelens. A high-pass filter with a blocking ratio of 108 for220 nm/193 nm transmission was interposed in front of thecamera to reject elastic scattering or parasitic reflectionsof the laser. Before each run, Rayleigh scattering picureswere registered (without high-pass filter) to obtain the meanirradiance field over the laser sheet, then O2 fluorescenceimages from a side jet of heated oxygen were registered tocalibrate the global sensitivity β(TR) of the imaging systemat the reference temperature TR = 800 K.

Figure 16 shows an example of a single shot imageof oxygen fluorescence in the reacting flow. As expectedno fluorescence is detected from the liquid dart of injectedoxygen which vaporizes quickly. Owing to the highefficiency of O2 LIF in the hot reactive zone (withoutprior cumulated laser absorption) the thin reaction surfaceis depicted by the sharp rise of the LIF signal when the

laser sheet reaches it. The detectivity limit due to thebackground noise level of the imaging system is about 2%of oxygen mass fraction in the reaction zone. After thissharp rise at the reaction surface, the radial evolution ofthe fluorescence signal toward the burner axis results froma complex competition between |O2| increase, temperaturedecrease and laser irradiance decay due to cumulativeabsorption.

On each of the 500 registered images, the radial gradientof the fluorescence signal was locally measured at inflectionalong with the local angle of the contour with the burner axis.The azimuthal component of the gradient was approximatedby the mean of its axial component at the same section.Figure 17 shows the derived map of the mean oxygenconsumption rate of oxygen in the turbulent flow. Closeto the injector the reaction rate exhibits a bimodal structurethat could be due to instabilities. The axial growth of thecumulated oxygen consumption was then derived by spatialintegration of the planar map of mean reaction rate andwith assumption of mean axisymmetry. The results indicatethat about 40% of the injected mass flow rate of oxygen isconsumed over 50 mm before the flame reaches the walls ofthe burner.

5. Conclusions

Planar laser induced fluorescence pictures of a given speciescan be converted into concentrations maps provided that asuitable configuration is used to minimize laser absorptionand fluorescence trapping biases and collisional quenchingdependence. The accuracy of the diagnostic in term ofabsolute concentration data is dependent on the precision

900

Applications of PLIF

of the calibration procedure for the global imaging system,whereas uncertainties on the relative spatial distributions stilldepend on the validity of the assumptions and calculationsfor the collisional quenching behaviour.

Examples of imaging applications have been givenfor OH, CH, O2 and temperature fields in turbulentreacting flows. These examples are far from constitutingan exhaustive review; fluorescence of NO and othermajor pollutant species are missing. Emphasis has beenput on the great potential of planar LIF diagnostic toinvestigate instantaneous conditional data (concentration andgradient, concentration and temperature, concentration andconcentration, concentration and velocity fields,. . .). Thesecorrelated space resolved data are useful and allow relevantinterpretations of an experiment with a deeper access to theinteraction phenomena involved in turbulent reacting flows.

References

[1] Eckbreth A C 1998 Laser diagnostics for combustiontemperature and species Energy and Engineering SciencesSeries vol 7 (Abacus)

[2] Daily J W 1977 Appl. Opt. 16 568[3] Stepowski D and Cottereau M J 1985 Prog. Astronaut.

Aeronaut. 95 642–57[4] Stepowski D 1992 Prog. Energy Combust. Sci. 18 463–91[5] Stepowski D and Cottereau M J 1979 Appl. Opt. 18 354[6] Stepowski D and Cottereau M J 1981 Combust. Flame 40 65[7] Crosley D R 1981 Optics Eng. 20 511[8] Cattolica R J 1982 Combust. Flame 44 43–59[9] Cattolica R J and Vosen S R 1984 20th Symp. on Combustion

(Pittsburgh, PA: Combustion Institute) p 1273[10] Allen M G and Hanson R K 1986 21st Symp. on Combustion

(Pittsburgh, PA: Combustion Institute) p 1755[11] Schefer R W, Namazian M and Kelly J 1990 23rd Symp. on

Combustion (Pittsburgh, PA: Combustion Institute) p 669[12] Smith G P and Crosley D R 1983 Appl. Opt. 22 1428[13] Cheng T S, Chao Y C, Wu D C, Yuan T, Lu C C, Cheng C R

and Chang J M 1998 J. Flow Visualization ImageProcessing 5 239–47

[14] Cheng T S, Wehrmeyer J A and Pitz R W 1981 29thAerospace Science Meeting (Reno, NV) AIAA Paper91-0181

[15] Lozano A, Yip B and Hanson R K 1992 Exp. Fluids 18369–76

[16] Fairchild P W, Smith G P and Crosley D R 1983 J. Chem.Phys. 79 1795

[17] Kohse-Hoinghaus K 1994 Prog. Energy Combust. Sci. 20203–79

[18] Cattolica R J, Stepowski D, Puechberty D and Cottereau M J1984 J. Quant. Spectrosc. Radiat. Transfer 32 363

[19] Nguyen Q V and Paul Ph H 1996 26th Symp. on Combustion(Pittsburgh, PA: Combustion Institute) p 357

[20] Barlow R S, Dibble R W, Chen J Y and Lucht R P 1989 27thAerospace Sciences Meeting (Reno, NV, 1997) AIAAPaper) 890061

[21] Dyer M J and Crosley D R 1982 Opt. Lett. 7 382[22] Barlow R S, Dibble R W, Starner S H, Bilger R W,

Fourguette D C and Long M B 1990 28th AerospaceSciences Meeting (Reno, NV) AIAA Paper 90-0732

[23] Maurey C, Cessou A and Stepowski D 1998 9th Symp. onApplications of Laser Techniques to Fluid Mechanics(Lisbon)

[24] Vanquickenborne F and Vantiggelen A 1966 Combust.Flame 10 59–69

[25] Miake-Lye R C and Hammer J A 1988 22nd Symp. onCombustion (Pittsburgh, PA: Combustion Institute)pp 817–24

[26] Peters N and Williams F 1983 AIAA J. 21 423–9[27] Schefer R W, Namazian M, Filtopoulos E E J and Kelly J

1994 25th Symp. on Combustion (Pittsburgh, PA:Combustion Institute) pp 1223–31

[28] Kaplan C R, Oran E S and Baek 1994 25th Symp. onCombustion (Pittsburgh, PA: Combustion Institute)pp 1183–9

[29] Muniz L and Mungal M G 1997 Combust. Flame 111 16–31[30] Schefer R W and Goix P J 1998 Combust. Flame 112

559–74[31] Reboux J, Puechberty D and Dionnet F 1996 SAE Paper

961205[32] Medaerts F and Puechberty D 1998 SAE Paper 982524[33] Kohse-Hoinghaus K and Meier U 1994 Non-Intrusive

Combustion Diagnostics ed K Kuo and T Parr (New York:Begell) pp 53–64

[34] 1997 ZODIAC Project, Final Report, CEE Programme JouleIII JOF3-CT95-0025

[35] Guerre S, Bazile R and Stepowski D 1996 26th Symp. onCombustion (Pittsburgh, PA: Combustion Institute) p 105

[36] Bilger R W 1977 Combust. Flame 30 277–84[37] Lee M P and Hanson R K 1986 J. Quant. Spectrosc. Radiat.

Transfer 36 425–40[38] GDR 1075 (CNES, CNRS, ONERA, SEP-SNECMA) Actes

du 6eme Coll. SEP DGPL Foret de Vernon, BP 802,27202 Vernon Cedex, France

901